feat: session-native subagents runtime v1 (spawn/list/info/kill)#217
feat: session-native subagents runtime v1 (spawn/list/info/kill)#217
Conversation
|
Progress update: completed the requested 5 incremental commits on this PR. New commits:
Validation:
I’ll keep this PR as the single integration branch unless you want me to split follow-up stabilization into separate PRs. |
|
继续同一个 PR 完成了你提的下一阶段 3 项能力,已推送:
配置补充(已在同 PR 中):
验证: running 7 tests running 6 tests running 8 tests running 17 tests running 0 tests test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s 全量通过。 |
|
继续同一个 PR 完成了你提的下一阶段 3 项能力,已推送:
配置补充(已在同 PR 中):
验证: |
Summary
This PR replaces the legacy in-tool
sub_agentloop with a session-native asynchronous subagents runtime.What shipped in v1
sub_agenttool entirely.sessions_spawnsubagents_listsubagents_infosubagents_killsubagent_runspersistence in SQLite with lifecycle + token accounting.subagents.max_concurrentsubagents.max_active_per_chatsubagents.run_timeout_secssubagents.announce_to_chatkillone or all) via cooperative cancel checks.docs/rfcs/0005-subagents-runtime-v1.md.Why this direction
This is a practical and safer architecture than tool-embedded delegation:
It aligns with OpenClaw/Spacebot subagent direction while keeping MicroClaw implementation lean and production-friendly.
Validation
cargo test -qFollow-up (v2+)
Related: #205